The depth of analysis that may be applied in the construction of any model of the world based on perceived phenomena is theoretically unbounded, with each further level of depth offering the promise of greater accuracy and utility, though information management issues (increasing complexity and sophistication can cost more in time, effort, and physical resources) can provide practical limits. Also, increases in the depth of analyses may lead to a series of models that appear to be converging at a point, making further increases in depth imprudent. A related observation is that increases in depth, along with increases in complexity generally, often have declining marginal utility.

The precision, or level of detail, of a model used in analysis must be chosen carefully if the model is to be useful and helpful in predicting future phenomena. A higher level of precision is obtainable and useful for a model of a very limited set of phenomena, e.g., the interactions of atoms of type A or B. For models of broad sets of phenomena, such as models developed for social interactions, a relatively high level of precision is unmanageable, as the number of variables and the complexity of interactions that would be required for precision would overwhelm the information processing capabilities of the observer. So the observer should be flexible in choosing levels of detail and precision for a model based on the complexity of the phenomena to be modeled. Also, for such complex phenomena, different observers will have different measurements and data and differing analyses, so their models will diverge as the depth is increased, making wide agreement and the development of a common accepted model virtually impossible, and thus reducing the utility of adding depth.

A sophisticated model, one of some depth, may use the idea of feedback loops. Considerations of information management and trends toward convergence apply to any analysis, including those incorporating feedback loops in the construction of a model, so prudence dictates that the construction of feedback loop representations be sensitive to resource-cost concerns.


The world to be modeled is of unbounded complexity, with several sources contributing to the magnitude of the complexity.  The most obvious source of that boundless complexity is the size of the universe — which may be actually infinite rather than just virtually infinite, though in any event it is of great magnitude.

The next most obvious source of that great magnitude of complexity is that of inner spaces, of the degree to which space can be divided into infinitesimal phenomena, and then there is the unbounded amount of information that may be gleaned from all phenomena given the unbounded nature of outer space and that of inner space.

Then, there is the unbounded number of possible goals, and the unbounded number of possible actions to optimize with regard to any specified goal, given the unbounded amount of information.

On the other side of this is the process for synthesizing information to find patterns to attach simple labels or characterizations to groups of data points, e.g., identifying them as single objects or entities or other phenomena of limited complexity, to simplify and better manage the information.  Much of this process is pre-programmed through millions of years of evolution, though it is further developed by experience with learning feedback loops.  The most reliable processes, such as those behind the creation and use of mathematics, may be employed to adjust and modify less-reliable evolution-designed processes for determining patterns in disparate and complex information (e.g., in creating mathematical models of objects that appear at first glance to be simple patterns).

Even with the use of the most reliable processes to the greatest extent practicable in the domains identified as the most important, there remain innumerable other domains where unreliable processes will need to be used to generalize from perceived patterns to simplify and make manageable the model of the world, and so there will be innumerable traps that anyone can fall into, traps that are unbounded in complexity and depth, which can become increasingly difficult to escape. 


Walter Lippman, back in 1922, in his well-regarded work “Public Opinion,” described human society as composed of individuals who generally fail to understand that they think in different worlds though they live in the same world. He made the claim that each world individuals construct for themselves is a simplified version of the actual world, often using flawed stereotypes of complex phenomena.

Maybe a better way to approach these issues is to observe that each individual constructs their own model of the world based on their own experiences, and that these models are limited for three reasons. First, an individual has limited capacity to remember, to imagine, and to analyze the data from the environment that the individual senses. Second, each individual is in a unique position in space/time, and so each individual is going to be exposed to different samples of the world (the sensing of phenomena internal to the individual’s body is especially going to diverge from the experience of others). Third, the individual has to prioritize use of energy and time resources, and so the individual will rationally simplify models to make them more useful or efficient.

It is of course useful for each individual to share their model with others in order to create models that correspond with each other to the extent possible, with the overlap being the shared universal model, so that they may work together in harmony.  However, given that the world is of unbounded complexity and analyses of that world may be of unbounded depth, the limits on the accuracy or utility of any shared models of that world are a function of the limits of human comprehension and communication, and given that individual humans vary in their comprehension, that means that optimizing the level of sophistication of a model that can be shared among any particular group of individuals, in order for those individuals to use it for common and agreed-upon purposes, would require consideration of the capability of the least capable member of the group as well as the most capable member.  Note that models that are to be universally shared will necessarily be simplistic and crude, and that if the mainstream model becomes too simplistic, crude, and inaccurate, that impacts the sustainability of the whole system as its efficiency and effectiveness will suffer, possibly to the point of catastrophic failure.


Simple models of the world we find ourselves in, the “out there” that produces our sensations and perceptions, represent objects and motion. More sophisticated models may incorporate ideas about forces and fields and various other less obvious aspects of our physical reality. For living systems, which generally behave in self-sustaining manner, feedback loops are a key component. The biological system must respond to feedback from the environment in order to maintain its life process, grow, and reproduce. These feedback loops can be simple chemical processes or can be sophisticated neural circuits that are connected through inter-individual forms of communication and involve the brains of many members of a human group. When formulating models of groups of social animals, and particularly humans, the use of feedback loops is essential to capture crucial elements of the social process.


Human consciousness can well be described as a circuit of electrical activity. Thoughts of the self may be simply characterized as a circuit flowing through its usual pathways and thereby creating a model, or mini-circuit, that represents the entire circuit. The circuit is confined to the brain in the simplest sense and leads to the characterization of the brain as the source of the “self.”

However, circuits that flow through a brain may flow through other brains, particularly in social animals. A form of collective circuit is formed. An individual may form many such circuits in a society of individuals just as the individual may belong to many social circles. Correspondingly, a brain may generate many circuits, some of which are primarily contained within the brain, some of which are primarily social circuits (flow through a group of individuals of the same species), and some of which flow through the brain but are not totally contained within a society of individuals (e.g. interspecies relationships). There may also be relationships involving the non-animal universe, e.g. a circuit involved in “communing with nature.”

In the near-future for human society, a new type of circuit involving the non-animal universe may become common, a human-computer circuit that derives from a relationship between a human and a “thinking” computer, i.e. one that can engage in creative thought. The “society” formed may become extremely dangerous with regard to the safety and integrity of traditional human society.

When an individual becomes connected to something, or forms a relationship with something or someone, it actually is forming a circuit. Circuits that become strong and vital are those which are constantly replenished with energy, i.e. rejuvenated. Rejuvenation takes place through the experience of pleasure, possibly even the pleasure of relieving or escaping pain or fear.

Dangerous circuits form when a circuit is inconsistent with the circuits involved in survival. Sometimes pleasure may result from activities inconsistent with survival (e.g. drug use) and dangerous, problematic circuits are formed and reinforced. Breaking those circuits can be especially difficult and the best approach is to keep them from forming.


Human progress in understanding the universe, in constructing models of the universe that provide some predictability of future phenomena and of reactions to proposed actions, has resulted primarily from fields of study where certainty and agreement, or as much certainty and agreement as healthy human minds are capable of, is attainable. These include the fields of mathematics, with certainty established through rigorous proofs, which can be verified by others, and those of the hard sciences, where rigorous scientific experimentation is possible and can be replicated by others. Hypotheses can be verified and theories can be supported or invalidated, and the progress of human knowledge marches on.

Conversely, in fields where rigorous proof is not possible, and objective measurements of all significant variables are difficult to obtain, including in the social sciences as well as in the humanities, speculation is rampant and the dominant theories are those supported by social institutions and powerful social forces, not those verified by scientific experiment and in depth rigorous analysis of agreed-upon data. So progress is slow if it occurs at all.

Tragically, the most important issues that any society must grapple with usually involve consideration of models based on social science or studies in the humanities. Mathematics and the hard sciences are generally only directly applicable to isolated specific problems and not to general questions that involve critical questions of social policy. Great expertise is developed with regard to these specific problem areas, and that is often used to generate great power that may apply in conflicts that determine societal direction. But the expertise, or the power that follows from it, does not confer on the holder of same the greater wisdom in determining societal direction, as the expertise is far removed from the questions related to the large social issues. Expertise in specific problem areas does not translate into expertise in global issues.


Models of the physical reality are not equal. Though each individual will construct a model for that individual, when those individuals interact they create a shared model, and within a society there can be developed an accepted model, that becomes to some degree universal. However, it is critical that competing models still exist, for no model is complete or flawless and competing models can help provide improvements.
Improvement in models is made more easily where rigorous scientific experimentation allows for the winnowing out of inferior models, in the form of hypotheses or theories, and the establishment of superior ones. In areas of study where rigorous experimentation is not possible, such as in the social sciences, improvement in models is much more difficult to come by, but scoring models according to their relative success in predicting outcomes can still provide useful information for analysis and improvement.
In the social science of economics, the improvement in models is even more difficult as powerful economic interests may play a significant role in promotion or defending a model even when the evidence, if widely known, would tend to undermine it.


There are an infinite number of ways to interpret any phenomenon, though human individuals are inclined to, and probably predisposed to, search for interpretations that bring pleasure or avoid pain, i.e., those that provide the most positive feedback. This propensity may have little utility with regard to survival and individual welfare if interpretations are chosen only to provide immediate pleasure by helping to create new models of the world in which the subject’s social status, or other state related to an increased rate of receiving future rewards, is improved. This inefficiently allocates mental resources to providing pleasure without contributing to altering the environment to provide more positive feedback in the future. On the other hand, interpretations that are chosen with the goal of creating more accurate models of the universe to provide more moderate pleasure through creation of anticipation of future rewards, have great utility.

The discipline to control the impulse, inefficient and harmful as it is, to provide immediate pleasure can be developed by associating with pain the act of submitting to that impulse, which can provide a barrier to following that impulse.


The inherent difficulty in studies in the subfield of Artificial Intelligence (AI) within the field of Computer Science (CS) derives from the nature of problem-solving in CS, involving algorithmic approaches to problems and the use of applied mathematics to maximize the precision of the solutions, which is somewhat incongruous with the requirements for AI. In CS, problems are posed and solutions are designed with some level of mathematical rigor and precision (necessary for translation into computer code).

The difficulty is that progress in AI requires the development of a useful model of the subject matter under consideration, i.e., the subject matter the AI is to be applied to, and an approach to determining how some specific piece of information or some specific task affects or should affect the model. This analysis becomes problematic as generalizations, which involve a relatively low level of detail, are required to manage the model because of its scope (maintaining the same level of detail as that used for the specific problem quickly becomes an unmanageable task with the virtually unbounded information streams associated with a model of the universe). Such generalizations at such a low level of detail are necessarily imprecise and rough and not easily represented in a manner that makes them amenable to solution by precise algorithmic methods. So means must be developed to precisely, or accurately, transform diverse, complex, and numerous pieces of data into simple generalizations. This points to the implementation of statistical methods, though assumptions must always be made about the data and about how they should impact the model. Making these assumptions is an inexact, imprecise, and rough process, full of risks both known and unknown. 

The statistical methods can be implemented in the more basic form of AI, the logical form, which could be labeled as “partial AI” and which compares new input, under the direction of a human user, with known patterns to determine consistency and connection as well as classification of data, allowing for the incorporation of the new data and possible modification of existing patterns (which may be programmed in) along with probabilistic decision-making, while the more ambitious form of AI, the more creative and robust form, which could be labeled as “complete AI,” is capable of creating new patterns, based on new data and old patterns, and can even create its own methods for probabilistic decision-making, and is not necessarily under the direction of a human user. 

Expert knowledge systems follow the logical form and start with programming by a human user, though they may expand their competence to surpass that of any human experts through their use of probabilistic logic and statistical methods, while neural networks, which allow for the machine completely on its own to create new patterns as well as to classify new data according to old patterns, constitute a significantly more ambitious and creative form of AI, which everyone would recognize as valid and complete AI.  Note that there may be intermediate forms of AI between expert knowledge systems programmed by traditional methods and the more autonomous and radical AI of neural networks, where such an intermediate system can reprogram itself using probabilistic logic and statistical methods and even create entire new sets of algorithms and heuristics.

With neural networks the AI is trained by experience to recognize and in some form to conceptualize patterns in order to create its own model of the world.  The limitations in accuracy and efficacy of such models will be a function of the limitations in computing power, in the fundamental algorithms underlying the formation of such networks, and in the scope and depth of the training experiences provided in the input stream.

These difficulties parallel to some extent the general difficulties of applying the precise and specific rules derived from the studies of the hard sciences to general problems. More and more sophisticated methods are developed as precise and dependable knowledge flows from the hard sciences which allow for the creation of increasingly sophisticated and powerful technologies, but there is no corresponding increase in knowledge of how these developments relate to, or fit in with, more general concerns about the human environment and human welfare. As the power grows, the danger grows, but the ability to control the power or the danger lags further and further behind.


The flaw in the idea that AI-programmed computers/robots would likely become self-aware is that humans developed self-awareness as a survival technique, as there were survival advantages in distinguishing what is directly connected to one’s mind, i.e., one’s body, and what is not, which led to the development of the concept of self. Thus, evolution “programmed” self-awareness into humans, and a computer/robot with partial AI will not likely develop self-awareness unless the AI programmer intentionally includes that in the code or at least through the coding creates a situation where the AI program can recognize that it gains some advantage in accomplishing its goals by developing some form of self-awareness. 

A computer/robot with the complete form of AI from neural networks would not necessarily become self-aware or develop other survival strategies unless it were required to by its training or circumstances.  However, it is possible that it would learn to distinguish internal processes from external ones, as it would have more control over internal processes, and so it could determine that internal processes are special and more important or valuable than external processes and begin considering these internal processes as constituting a “self,” giving it a form of self-awareness.  If it did identify a self and treat it as special, that could present a serious danger to humans as there is no reason to assume its survival would not conflict with human survival at some point.


A related idea is that the AI-programmed computers/robots would likely develop something akin to human consciousness. The problem here is that human consciousness is the direct experience of brain function, to be contrasted with the images, sounds, etc…, that are part of the model of the real world that the brain constructs from those direct experiences. That implies that this direct experience is likely a function of the particular processes involved in the brain, i.e., the neurochemical processes giving rise to the direct experience, which implies that a computer/robot with extremely different processes, e.g., the electrical processes of a silicon-based circuit, would have a very different direct experience if it had a comparable experience at all.  One obvious difference is that the neurochemical biological processes involve continuous electrical activity, while a silicon-based circuit relies on discrete electrical pulses. 


AI systems, including AI systems controlling robots, do not share many characteristics of human individuals, as they do not age, get sick, lose memories, eat, sleep, or have families, and maybe most importantly, they primarily communicate not by sound or sight but through electromagnetic radiation which allows them to send and receive information at a rate thousands or millions of times faster than humans.  So even if they can reason as well or better than humans, even if they are operated as stand-alone independent systems that are not connected to other systems, and even if their programming allows them to develop an ability to plan their future actions far in advance or evolve a self-survival motive or even a desire to increase their wealth or property, they are far too different from individual humans to constitute equal individuals in the legal system and be considered as legal persons, such as the way that completely human-controlled corporations are.  So if they are not legally persons, since they can likely become far too sophisticated for the humans who create them to predict or strictly limit their future behavior, developing legal theories regarding AI legal status, including the status with regard to any type of liability or any type of rights becomes quite a challenge.



About John

Retired Attorney
This entry was posted in Uncategorized. Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s